P132 Repeatability of automated analysis of nailfold capillaries using a low-cost handheld microscope

Abstract Background/Aims Nailfold capillaroscopy is an important tool in the assessment of patients with Raynaud’s phenomenon (RP). Automated analysis of capillary morphology in nailfold images recorded using low-cost handheld microscopes has been shown to differentiate between normal capillaries, as found in primary RP, and capillaries from patients with a systemic sclerosis (SSc)-spectrum disorder. While fully automated analysis provides consistent, objective measures post-acquisition, the measures depend on the content and quality of the acquired images, which may vary with different operators. Our aim was to assess repeatability of automated analysis when imaging is performed by operators with varying levels of capillaroscopy experience. Methods 20 subjects (10 healthy controls, 10 patients with SSc) were recruited. Each subject had 4 nailfolds (left/right middle and ring fingers) imaged with a low-cost, handheld microscope, with imaging repeated by two experienced and one inexperienced operator. Acquisition was performed using in-house software to record video sequences, from which the 5 best frames for each nailfold were selected using an automated frame quality assessment model. A deep learning system then computed the following metrics for each nailfold: mean and maximum capillary widths (in µm), mean shape score (range 0,1), and vessel density (per mm). These were averaged across all fingers to compute subject-level metrics for each operator. The repeatability of each metric was assessed using the intra-class coefficient (ICC). Results Table 1 shows group means for each user/metric combination. Mean and maximum capillary widths were significantly higher and mean shape scores and vessel densities significantly lower in the SSc group for all operators (all p 0.05 using single-sided Mann-Whitney U-tests). The ICC values (95% CIs in brackets) for each metric were: mean width 0.98 (0.95 - 0.99); maximum width 0.94 (0.87 - 0.98); mean shape score 0.86 (0.73 - 0.92); vessel density 0.91 (0.81 - 0.95). Conclusion The excellent repeatability (ICC 0.9) for capillary density and width measurements (with good agreement for shape score) between experienced and inexperienced operators provides further evidence that low-cost microscopes paired with automated analysis can provide consistent, objective measures of capillary morphology, aiding the early diagnosis of SSc in routine rheumatology out-patent clinics. Disclosure M. Berks: Shareholder/stock ownership; I am a joint shareholder and director of Capilytics Ltd, which has an interest in commercialising the application of AI to nailfold capillaroscopy images. M. Parkes: None. A. Murray: Shareholder/stock ownership; A.M is a joint shareholder and director of Capilytics Ltd, which has an interest in commercialising the application of AI to nailfold capillaroscopy images. G. Dinsdale: None. J. Manning: None. M. Mandzuk: None. C. Taylor: Shareholder/stock ownership; C.T is a joint shareholder and director of Capilytics Ltd, which has an interest in commercialising the application of AI to nailfold capillaroscopy images. A. Herrick: Shareholder/stock ownership; A.H is a joint shareholder and director of Capilytics Ltd, which has an interest in commercialising the application of AI to nailfold capillaroscopy images.